Solaria-1

Solaria-1 is the model on which the Paris company Gladia built its business, announced on 3 April 2025. It remains on sale after the release of Solaria-3 in June 2026, and the manufacturer positions the two as complements rather than as a replacement pair — an unusual arrangement worth reading carefully, because for several kinds of audio the older model is still the one the company recommends. Its distinguishing claim is coverage. Gladia says the model handles 100 languages, of which 42 are supported by no other commercial transcription API, and that it switches between them inside a single conversation without configuration. That is the opposite of the trade its successor made: Solaria-3 is tuned for five European languages and describes its own code switching as limited. Solaria-1 is also the only one of the two that transcribes live audio. The company reports partial results in under 103 milliseconds on the model page — worth noting alongside the 270 milliseconds quoted in its own launch article, a discrepancy the manufacturer does not explain, and one that most likely reflects a different point in the pipeline being measured rather than a threefold speed-up. On accuracy it is now behind its successor almost everywhere: 12.9 percent word error rate against 9.6 on Gladia's annotated English customer calls, and 8.1 against 6.4 on Earnings22, a set of financial conference calls. But it wins on two public benchmarks made of clean, formal speech — parliamentary recordings and read audiobooks — and Gladia tells customers with such material to stay here. Pricing is the same as for the newer model: 0.61 dollars per hour of recorded audio and 0.75 dollars per hour of live streaming on pay-as-you-go, falling to 0.20 and 0.25 dollars on contract. The model is closed, sold as a hosted service, and available as an on-premise deployment. All comparative figures are the manufacturer's own.

▸Videos

No videos yet.

▸Reviews